Vision Based Topological State Recognition for Deformable Linear Object Untangling Conducted in Unknown Background

Abstract

In this paper, we propose a deep learning based method to recognize the topological state of a deformable linear object (DLO). The utilization of deep learning can ensure that topological state recognition is robust to background change. This feature is useful if applications of DLO manipulation in real environment. And this feature has never be realized. In addition, the proposed scheme is also applied to the situation when multiple DLOs exist. This situation has never been considered. By integrating the proposed topological state recognition method and DLO untangling strategy, rope untangling experiments are conducted for both the situations of containing a single DLO and double DLOs.